Assignment 3 Solutions
Assignment 3 Solutions
Greg Parker
September 27, 2015
It is easier to converge in the weaker topologies, so convergences implies convergence for the topologies
to the left, and failure to converge implies failure to converge for the topologies on the right.
i) f (t), h(t), g(t) are all continuous in the product topology because the projection maps into each coor-
dinate are all continuous (by Theorem 19.6).
ii) g, h are continuous in Q the uniform topology. It suffices to show this on a basis. For g, a basis element
around an image point is (t , t + ) whose pre-image is (t , t + ) which is open. For h, it is simpler
to use to metric definition. Pick > 0, and let = . Then if |y x| < , we have | n1 x n1 y| < in each
component, so the supremum is < as well.
f however is not continuous in this topology. If we consider (1, 1) in each copy of R this is open in the
uniform topology, but the pre-image is ( 1 1
T
n n = 0 which is not open.
, )
iii) None of the functions are continuous in the box topology. Non continuity of f follows from the
uniform topology case. For g consider the open set (1/n, 1/n). As above, the pre-image is {0}. For h,
consider (1/n2 , 1/n2 ) and the pre-image is again {0}.
Q
(b)
Convergence follows the same implications as continuity.
i) wi {0} in the product topology. Take an open set around 0. It is of the form U1 ... Un R...
with each Ui containing 0. Then for j > n, wj is contained in U .
wi is not convergent in the uniform topology. Consider B1 ({0}). wi is not contained in this for any i.
Hence wi is not convergent in the box topology either.
ii) xi 0 in the uniform topology. Consider B (0). There exists an N so that N1 < . Then for j > N
we have xi B (0). Therefore xi converges in the product topology as well. Q
In the box topology however, {0} is the only reasonable possibility for a limit, but the open set (1/(n+
1), 1/(n + 1)) contains no xi . So xi does not converge in the box topology.
iii) yi 0 in the uniform topology, because for > 0 there is an N so N1 < and then for j > N we
have yi B (0). Therefore we have
Q convergence in the product topology as well.
In the box topology, the set (1/(n + 1)2 , 1/(n + 1)2 ) will never contain points of yi .
iv) zi 0 in the box topology, so in all topologies. Pick an open set U = U1 U2 ... we can reduce to
intervals in the first two components and take the minimum width so (, ) (, ) U3 ... U , then
for n > 1/e we have zn U .
1
2 Munkres Section 20 Problem 8
Define a topology on ( rInI )
i=1 R by defining the distance of two sequences {xi }, {yi } as
I
sX
min 1, (xi yi )2
i
(a)
We have the inclusion of topologies
Then we would have B/ ({x})2 B ({x}) as if d({x}, {y})2 < / then d({x}, {y}) < d({x}, {y})2 < .
And we have shown inclusion of element of the base. However = 1 suffices as
X
(xi yi )2max (xi yi )2
i
as the maximum term is obviously included in the sum. Taking the square root gives (2.1)
ii) Now we show a set open in the `2 topology is open in the box qtopology. Take
q B ({x})2 . We find an
1 2 1 2
Q
open set around x in the box topology contained in it. Take (xi 2 2n2 , xi + 2 2n 2 ). A point {yi } has
2 X
2
2
(xi yi )2 < (x i y i )2
< < 2 d({xi }, {yi })2 <
2n2 i
2 6
(b)
Again, Theorem 20.4 says the topologies excluding `2 are distinct. Hence we need only show `2 is indeed
distinct.
We exhibit an open set in the box topology that is no open in `2 . Consider ( n1 , n1 ). If there were
Q
2
some open ball in ` topology in it, it would have to be center around 0. Yet B (0) contains {0, ...., , 0, ...}
for in every place, but for large n this is eventually not in the above product.
We now exhibit an open set in the `2 topology not open in the uniform topology. Take B ({xi })2 , and
suppose there were some so that B ({xi }) we P contained in the first ball. However, we can take {xi +/2},
which is in the -ball, but not in the `2 ball, as i 2 /4 = , a contradiction to the existence of such a .
(c)
We now consider the Hilbert Cube Y 1
[0, ]
n
in the subspace topology.
We have
box `2 = Uniform = Product
2
Proof. First we show the box topology is distinct still. Consider U = (0, n1 ) which is open in the box
Q
1
topology. Then consider the point {xi } = (n+1) U . If U were open in the `2 topology we would have
1
> 0 such that B ({xi })2 U . However, for sufficiently large n we have /2 > n(n+1) so for such an n,
2
{yi } = {xi + i,n /2} is contained in the ` ball, but not in U , as in that interval, Ui contains points only
1
n(n+1) greater than x. Note we may possibly have to reduce so it is still in the cube, depending on how
we choose n.
It suffices to show that the Product Topology is finer than the `2 topology on the cube. This gives
inclusions the other direction hence equality. Consider a ball P B ({xi })2 . We construct a set open in the
product topology contained in the ball. Let N be so large that N n12 < 2 /2. Then choose so small that
2 N < 2 /2. Then consider
U := (x1 /2, x1 + /2) (x2 /2, x2 + /2) ... (xN /2, xN + /2) R...
d(x, y)
d0 (x, y) =
1 + d(x, y)
Then (X, d0 ) is a metric space with the same topology, and a metric bounded by 1.
Proof. We must show that d0 is a metric if it is such, then it is obviously as d > 0 so d0 < 1. d0 is symmetric
and positive unless x = y because d is. To show the triangle inequality is satisfied, observe that the function
x
f (x) = 1+x has f 0 = (1+x)
1
2 and so x < y f (x) < f (y).
Then
d(x, y) d(x, z) + d(z, y) d(x, z) d(z, y) d(x, z) d(z, y)
d0 (x, y) = < = + +
1 + d(x, y 1 + d(x, z) + d(z, y) 1 + d(x, z) + d(z, y) 1 + d(x, z) + d(z, y) 1 + d(x, z) 1 + d(z, y)
where the last inequality follows from positivity of d because removing a positive term from the denominator
can only make it larger. The right hand side is exactly d0 (x, z) + d0 (z, y). Hence we have the triangle
inequality.
Now I claim the topologies coincide. To show this it suffices to fit a d0 ball inside a d ball and vice versa.
So let B (x) be a ball in the d metric. I claim that B (x)0 in the d0 metric is contained in it for = 1+
.
d(x,z)
Consider z B (x)0 then d0 (x, z) = 1+d(x,z)
< 1+ . Then by the strict montonicity of f , we must have
d(x, z) < as desired. Conversely, for a ball of radius in the d0 metric, B (x)0 , then B (x) for sufficiently
small that 1+ < is contained in it, as if d(x, z) < then d0 (x, z) < 1+ again by montonicity of f .
3
I claim fn (x) f (x) for every x. Obviously this is the case for x = 1 because fn (1) = 1n = 1 for all n so
the sequences is constant. If x < 1 then for fixed x we have fn (x) = xn which is a monotonically decreasing
function of n converging to 0 as we can write it e ln(1/x)n for 1/x > 1.
However the series does not converge uniformly, as the limit (which is unique as it is unique for each x),
is not continuous, hence uniform converge would contradict Theorem 21.6.
(b)
For A X define a retraction X A as a continuous map that fixes each point of a. Then r : X A a
retraction is also a quotient map.
Proof. By (a) it is enough to show the existence of a right inverse A X, but the inclusion map suffices.
4
To show q is not closed, we find a closed set in A whose projection is not closed. Consider the graph of
1
x. A graph of a continuous is always closed (use limit point definition), and the function is only defined for
x > 0. Hence the projection to the x-axis is (0, ) R which is not closed.
(b)
Consider now R2 under the equivalence (x0 , y0 ) (x1 , y1 ) if x20 + y02 = x21 + y12
it is homeomorphic to [0, ).
Proof. Equivalence classes are circles of any radius, including 0. The proof proceeds in an identical fashion
to part (a): We define a map on equivalence classes by [(x, y)] 7 x2 +y 2 . This is well-defined by construction
of the equivalence relation, and injective and surjective for the same reasons as above. The inverse image of
(a, b) under this map and the projection is either the open disk or the open annulus, hence its continuous.
And an open set likewise pushes forward to an open set.
sends
(x, y) (x, y 1 ) x y 1
and it is the composition of continuous maps, as it is easy to show that id f for f continuous is
continuous as a map out of the product topology.
Now suppose m given by (x, y) 7 x y 1 is continuous. Then
i m
H H H H
by
y (id, y) id y 1 = y 1
5
is continuous, as the inclusion of H into H H by id in the first coordinate is continuous (inverse image of
open U V is simply V which is open). Thus weve written inversion as a composition.
For the product, consider
id1 m
H H H H H
(x, y) (x, y 1 ) x (y 1 )1 = xy
is continuous as it is the composition of continuous maps, including the inversion map just constructed.
(, ) ei()
Non-Book Problems
11 Problem 1
This problem is identical to problem 8b.
12 Problem 2
Define C1 = [0, 1/3] [2, 3/1] and define Cn iteratively by removing the middle third of each interval. Then
define the cantor set as
\
C= Cn
n=1
Alternatively, define C as those points in [0, 1] that have ternary decimal expansions containing only 0s
and 2s.
6
(a)
C is closed.
Proof. Each Cn is closed, and C is therefore the infinite intersection of closed subsets, which is closed (use
the fact that infinite union of open sets is open and take complements).
(b)
C is a perfect set. That is, for each c C there is a sequence xn C so that xn 6= c for any n and xn c.
Proof. Since c C, c Cn for each n. Each Cn consists of several intervals of length 31n . In particular, c
falls in one of these for each n. Let xn be the upper endpoint of the interval into which c falls in Cn . By
construction, each xn is within 31n of x, and for j > n xj is closer (or equally close) to c than xn , as it is an
endpoint of a subinterval. Hence xn c.
The end points of each interval, are of course in C because they have ternary expansions ending in infinte
sequences of 0s or 2s.
(c)
The interior of C is empty. i.e. for no c C is B (c) C for any . Equivalently, for every c C we can find
a sequence yn / C so that yn c.
(d)
C is uncountable.
Proof. Notice that if for X uncountable f : X C is injective for all but countably many points, then
C is uncountable because we have found an injection from an uncountable set. In particular card(C) >
card(Im(f ))
To this end, define f : [0, 1] C by f (1) = 1 and for all other points, x [0, 1) write x = a1 a2 ... in the
binary expansion with the convention that we may not have an infinite string of 1s, and change all 1s to 2s.
This gives a point in C. I claim this map is injective at all but countably many points. Consider two points
x, y [0, 1], they map to different decimal expansions in C, so the only way they can be the same point is
if those decimal expansions are two different expansions for the same number. This can only happen if the
image point in C ends with an infinite string of 0s or 2s, but this would require one of the point in x, y to
end in an infinite string of 1s, which is disallowed by our convention. Thus all point not ending in an infinite
string of 2s or 0s are in the image of f . This set not in the image is countable, as each is the endpoint of an
interval in some Cn , hence rational (and Q is countable).
References